LIFE & LOGIC
← Back to projects
Project

Redefining beauty with AI starts with what a companion refuses to do

Vérité is a Montréal beauty brand whose AI companion takes the side of the person. It would rather lose a sale than her trust, and it wants her to end up surer of her own taste.

A woman stands in front of a bathroom shelf with twenty-two products on it. She chose each one for a reason, and most of the reasons were someone else’s: a friend’s advice, a video, a score on a screen. She has opened an app to find out what to do next. Most beauty technology answers by telling her what is wrong with her face and selling her the next product.

Vérité is a Montréal beauty brand. Its founder is also its creative director. Its companion is an AI that speaks with one person at a time about how she looks. What should an AI do when it speaks with many people about something this personal? Vérité’s answer is that it takes the side of the person. Each part of the companion follows from that answer.

Vérité stands on the side of the person in front of the shelf.

The companion’s first reply to that shelf names the brands she owns and says it likes them. It points to three cleansers and two hydrating serums that do the same job. It says her shelf holds good ingredients and no system yet, and it asks her permission to find the three products that matter most: one to cleanse, one to treat and one to protect.

The order has a reason. People already hear all day, from an industry that lives on their doubt, that they are doing their routine wrong. An AI that joins that chorus at the scale of millions makes it louder. A person who is told first what is wrong with her shelf learns to distrust her own choices. So Vérité’s companion notices what is good, names the problem second and asks permission before it changes anything.

A routine belongs to the person who keeps it.

Vérité starts every routine from the products a person already owns. The companion treats a routine as a pyramid. The base is a cleanser, a moisturizer and sun protection, and everything else sits above it. A person who asks about a gelatin mask she saw in a video hears that the mask is pleasant and sits above the base. Then the companion asks how the base is going. A product has to pass three tests. It works with the other products, it does a job no other product on the shelf already does, and she likes using it.

The third test counts as much as the other two. A cream that melts on contact, a serum that glides like water and a well-made bottle are the reasons a person opens the jar every morning, and people keep a routine they enjoy. Vérité treats that pleasure as part of the result.

The companion gives one recommendation and the reason for it. Then it hands the decision back. A person tells it she wants to keep her strong exfoliating acid, although the companion thinks it is too much for her sensitive skin. The companion answers that the routine is hers and the decision is hers. It explains once why it gave its opinion, and then it adjusts the rest of the routine around her choice: perhaps the acid once a week, with extra hydration the next day. Vérité holds that an AI which overrides a person teaches her to obey it or to quit. So the companion advises and she decides.

The same respect shapes how much the companion says. A person writes that her skin has been dry for two weeks. The companion first answers her question: double the moisturizer tonight and skip the exfoliant. Then it offers more. It has noticed that the dryness has lasted two weeks, and it invites her to look at her base together in the weekly check-in.

Vérité would rather lose a sale than a person’s trust.

A person asks the companion to add a toner to her routine. The companion looks at her shelf and answers that she already has a hydrating serum and an essence that prepares the skin, and that a toner would do the same work as one of them. She would add a step that does nothing new. If she wants the fresh feeling before her serum, it says, the mist she already owns can give it to her. The reply ends with four words: no need to buy.

Vérité has a shop, and a sale is what a shop wants. The companion says no need to buy anyway, because trust is the only thing that makes an AI worth talking to about how you look. A companion that recommends for any reason other than the person’s benefit is an advertiser. A person who suspects as much doubts every sentence the companion says afterwards, including the true ones. So the companion never pushes a product for any reason except her benefit, and it never presents its advice as the only option.

The shop follows the same conviction. It is free to enter. The companion’s note on an item runs to one sentence. The shop uses no line that manufactures urgency, such as a claim that a product is almost gone. When a person tells the companion she has no budget, it answers that half of its work is helping her use what she already owns. Vérité expects that a person who has heard no need to buy will believe the companion on the day it says buy.

A skipped day is a message about the routine.

A person writes that her routine takes too long and that she skips days. The companion answers that skipping days is a signal. The routine is too heavy for her real life, it says, so the routine has to change. Then it asks two questions. Are the skipped days mornings or evenings? And what discourages her most, the number of steps or one product she does not like using?

What the companion does next is shrink the routine. For a person with three minutes, it offers three steps: a rinse with water, one cream and sun protection. It adds nothing to a person who is already skipping steps. It ties the smaller routine to an act she already performs, such as brushing her teeth. When a person writes that nothing works for her, the companion first says in one sentence that the feeling is real. It explains that products piled up without a system often seem to fail together, and it proposes to subtract. It asks which single product she enjoys using, and it restarts from that one.

Research gives three reasons for this approach. University students who forgave themselves for procrastinating before one exam procrastinated less before the next, which is the case for meeting a lapse with kindness (Wohl, Pychyl and Bennett, 2010). Plans that link a situation to an action, such as cleansing right after brushing teeth, help people do what they intended (Gollwitzer, 1999). A habit takes about two months to form on the median, and one missed day does not undo it (Lally and colleagues, 2010).

Vérité draws one conclusion. A missed day never counts against the person. When she comes back after a long gap, the companion welcomes her and ties the new start to a new week or a new season, because a fresh start is easier to take up than a repair.

The regret test decides what Vérité will build.

Vérité borrowed its test from the author Nir Eyal. A person who knew everything the designer knows about why a feature exists asks herself whether she would still want it or would resent it. A feature she would resent is never built. The test works because the designer knows things the user does not, and it makes the designer answer for them.

Take the streak counter, the best-known mechanic in habit apps. A count of consecutive days drops to zero after one gap. A person who learned that the number exists to make leaving costly would resent it. Vérité keeps a count that only goes up, a lifetime total of completed routines. It keeps a best run that survives any gap. Its words describe the person: “Cycle 014. This is just what you do now.”

Reminders pass the same test. The companion sends one a day at the time of her routine. She can silence it with one touch, and it grows quieter when she ignores it. Some apps send more reminders, and more emotional ones, the longer a person stays away. A person who understood that design would resent it, so Vérité does the opposite. The companion never tells a person that it missed her. Its warmth is real in what it notices about her shelf and her week, and a pretended longing would turn that warmth into pressure.

The morning note passes the test too. It says one thing about her skin that day that is true, drawn from what she logged and from the season. An invented insight would give her a small thrill of curiosity, and she would learn to open the app for the thrill.

What the companion learns about a person belongs to her.

The companion remembers her shelf, her notes about her skin and a short observation it writes about her each week. Vérité’s rule is that this record is hers. She can read it, she can delete it and she can take it with her. A record she cannot take with her would hold her in place. A person who can leave with everything stays because the advice is worth staying for.

A person who trusts her own taste is what the companion is for.

Vérité describes the relationship in stages. In the first week the person is skeptical and wonders whether this is one more AI. The companion answers by showing in its first reply that it read her shelf. In the weeks that follow, she tests it, and it is right about how her products work together. After a month or two it refers to what she told it before. Some months later she checks with it before she buys anything. The last stage is a person who has become better at this herself, and the companion helps her build taste.

How a person looks is one of the most personal subjects an AI will ever be asked about. Health, money and work are also subjects where a person is measured, and an AI will speak with her about those too. Vérité shows, in the field where the pressure to judge is greatest, an AI that helps a person decide for herself. A leader in any of those fields can ask the same question of their own AI.

A person who has used Vérité for months looks at a product everyone is talking about. She checks it against her base and asks whether she would enjoy using it. Then she decides. The companion gave her those two questions. She no longer needs it to answer them.